Retinopathy

DIABETIC RETINOPATHY USING SUPPORT VECTOR

MACHINES

Submitted By:

Prateek Goyal 10102219Ayush Garg 10104660

Introduction

Diabetes is one of the primary causes of causing blindness, almost 5% of the total cases in the world can be attributed to it.

For a proper cure, a proper diagnosis is also very necessary.

Due to this major concern, we provide an automated system to robustly detect diabetic retinopathy from the fundus images thereby providing a chance to save the vision of many people.

• To provide, the vast amount of population affected by diabetic retinopathy a proper diagnosis.

• Need to provide a cheap and robust solution in the field.

Why this domain ???

People face the following problems:Expensive DiagnosticNo availability of such techniques in the eye clinics.Testing must correspond to the strict regulations in the medical practices.Hypertension can onset retinopathy, so that must also be taken into account.

Why consider diabetic retinopathy ?

We are living in the world where modernization is bringing non-invasive techniques with the help of computer science and technology for diagnosis of diseases

What is diabetic retinopathy ?

Diabetic retinopathy (DR) is an eye disease caused by the complication of diabetes and we should detect it early for effective treatment. DR can be a complication of diabetes type 1 or diabetes type 2.

As diabetes progresses, the vision of a patient may start deteriorate and lead to diabetic retinopathy.

• DIARETDB1 diabetic retinopathy database and evaluation protocol. (TOMI KAUPPI, VALENTINA KALESNYKIENE, ET AL., 2011)

• SVM and Neural Network based Diagnosis of Diabetic Retinopathy. (R.Priya, P. Aruna, 2012)

• Object Recognition and Detection by a Combination of Support Vector Machine and Rotation Invariant Phase Only Correlation

(CHIKAHITO NAKAJIMA, NORIHIKO ITOH, ET AL. , 2000).

Previous work

We utilize a standard set of fundus images to train our SVM to detect DR with respect to the 4 features present during the ailment.

Since this field is a current research one, we take the opportunity to categorize a significant part of the existing work in a way that can help doctors in making the right decision.

we will deal with the following techniques

Solution to the big problem !!!

Image Processing

Feature Classification

Support Vector Machines

Feature Extraction

• Color Eye Image: The original fundus image.• Split Image into individual channels• Adaptive Histogram: Histogram equalization highlighting the darker regions and keeping

the brighter regions at almost same level.• Median Blur to remove noisy elements.• Thresholding to implement spatial filters.• Hough Circle Transform• Morphological Closing: Dilation followed by an erosion, to close the gaps hiding random

features.• SVM classifier

Implementation

Components

1. Input – Input is an image of RGB format.

2. Label – Final result obtained from the input image. 0 indicates disease is not there,1 indicates person is suffering from the disease.

3. Feature extractor- It takes input and applies the features to it.

4. Machine learning Algorithm- Supervised algorithms like SVM which take features as input.

5. Classifier Model- It is a model which actually predicts the output based on the training set which was earlier fed to the SVM.

Feature Extractor - It takes “Input” i.e. image(s) as input and creates a set of feature List for each image.

Before sending the images to feature extractor they must be preprocessed

Gray scaleHistogram equalization

Discrete wavelet transformation

Filtering to remove noise/unwanted components

Edge detection

Feature Extractor

Feature Extractor

Feature extractor gives a feature vector. It is a list of features.

In our case these features come out to be the problems arising in the retina of the eye:-• Hard Exudates• Soft Exudates• Micro Aneurysms• Hemorrhages

Machine Learning Algorithms

A supervised machine learning algorithm analyzes the training data and produces an inferred function, which can be used for mapping new examples.

Supervised Machine Learning Algorithm

A Support Vector Machine (SVM) is a discriminative classifier formally defined by a separating hyperplane. In other words, given labeled training data (supervised learning), the algorithm outputs an optimal hyperplane which categorizes new examples.

OBTAINED ACCURACY:

Classifiers Accuracy

Support Vector Machines 95.6%

PCNN 89.6%

• This result is obtained from SVM and Neural Network based Diagnosis of Diabetic Retinopathy.

(R.Priya, P. Aruna, 2012)

Therefore we use SVM in our predictions as in medical cases accuracy is our prime concern. Thus, providing us the final accuracy of >95%.

THANK YOU !!!